Applying Machine Learning to Predict Whether Learners Will Start a MOOC After Initial Registration

Online learning has developed rapidly in the past decade, leading to increased scientific interest in e-learning environments. Specifically, Massive Open Online Courses (MOOCs) attract a large number of people with respective enrollments meeting an exponential growth during the COVID-19 pandemic. However, only a small number of enrolled learners successfully complete their studies creating an interest in early prediction of dropout. This paper presents the findings of a study conducted during a MOOC for smart city professionals, in which we analyzed demographic and personal information on their own and in tandem with a small set of interaction data between learners and the MOOC, in order to identify factors influencing the decision of starting the MOOC or not. We also applied different models for predicting whether a person previously registered to a MOOC will eventually start it or not, as well as for identifying the most informative attributes for the prediction process. Results show that prediction reached 85% accuracy based only on the number of the first days’ logins in the MOOC and few demographic data such as current job role or occupation and number of study hours that the learner estimates he/she can devote on a weekly basis. This information can be exploited by MOOC providers to implement learner engagement strategies in a timely fashion.